BLOCKCHAIN AI.NEWS

Infrastructure

A Grafana Test Instance With Default Admin Credentials Was Chainstack’s Way In

The node provider’s November 3 notice said the attacker came “through a third-party component.” A technical section added November 10 describes something else: a test dashboard left on the internet since July, an MCP tool server with no login, and an intrusion tool that runs LLM agents.

✓ Disclosed by Chainstack (created Nov 3, technical details added Nov 10) · First outside coverage: The Hacker News (Nov 3) · This desk read Chainstack’s page directly; every timestamp below is Chainstack’s, not independently verified

Chainstack, which sells blockchain node access to developers, published an incident report on November 3. It said an external attacker “gained unauthorized access to part of Chainstack’s infrastructure through a third-party component,” was detected and cut off, and that customer nodes, the console and the API were not modified. The Hacker News ran that wording the same day. The page carries a later addition, dated November 10, that tells a more specific story, and it is worth reading against the first paragraph.

What the technical section says

According to Chainstack’s page, a test Grafana instance was deployed to a development Kubernetes cluster in July. It was reachable from the internet with no single sign-on or basic authentication in front of it, and its default administrator credentials “had never been changed.” The company states that it found no evidence a software vulnerability was exploited. Its summary of the chain is “default credentials, exposed administrative functionality, missing authentication, over-broad permissions and long-lived credentials.”

That is a description of Chainstack’s own configuration. Grafana was a third-party product, so the overview is not false. But the technical section does not point at a vendor’s flaw; it points at how the component was deployed and left.

The chain, as Chainstack describes it (UTC)

WhenStep
JulyTest Grafana deployed, default admin credentials, no login layer in front
Oct 26Attacker logs in, uses Grafana’s data-source proxy to reach internal services; reaches an MCP tool server with no authentication and cluster-admin rights
Oct 29Workflow-automation platform rewritten to run shell commands; machine identities stolen; first production secrets read
Oct 31Bulk reads of the production secrets manager; VPN key used to reach the database network; production databases queried; intrusion detected and cut
Nov 2Last observed use of a copied credential; investigation closed
Source: Chainstack incident report, technical details section. Condensed by this desk.

The tool server, and the tool that drove it

Per the report, Grafana’s data-source proxy let an administrator point the server at any internal URL, and no destination allowlist or network policy limited where it could connect. Over about two hours the attacker reached dozens of internal services. One was an internal Model Context Protocol (MCP) tool server that exposes Kubernetes operations to AI agents. Chainstack says it required no authentication and ran under a service account bound to cluster-admin. The company says the individual tool calls could not be recovered, but network flows and the changes that followed indicate it served as the attacker’s command executor.

Chainstack also says it assesses “with high confidence” that the attacker used ARTEX, an open-source framework that orchestrates LLM agents to automate intrusions, citing naming markers in the attacker’s resources and the reporting of Google Threat Intelligence, CrowdStrike and S-RM on ARTEX. It is careful about the limits. Public reporting ties ARTEX to several unrelated groups, so the company says it attributes the tooling, not the actor. The Hacker News, by contrast, reported overlap with a China-linked cluster; that claim comes from the article and its sources, not from Chainstack.

How it was caught

By a side effect, on Chainstack’s account. The attacker used a stolen private key for a VPN peer on the database network. Because two clients then presented the same peer identity, the tunnel endpoint flapped between them and a database replication pipeline failed. That failure started the investigation on October 31. The access paths were cut that day, but Chainstack says some copied credentials, such as API tokens for external services, kept working until rotated; the last use it saw was at 06:28 on November 2. The report also says two development hosts briefly brought back up that morning re-ran the attacker’s implants, with outbound connections blocked.

What customers were told, and what is unclear

Chainstack says a database of customer account information was accessed: organization names, account holder names, member email addresses and password hashes. It does not say how many customers or users are affected, and it does not say which hashing scheme the customer passwords use. Systems holding node credentials were also reachable; the company has no evidence they were read but is treating them as exposed. Customers are asked to change passwords (including reused ones), rotate node credentials, regenerate platform API keys and watch for phishing. Self-service rotation is live, and current endpoints keep working until customers finish.

One detail is easy to misread. The technical section says the attacker queried PostgreSQL’s pg_shadow view on one instance, which holds database role password hashes. That is a separate item from the customer password hashes in the overview, and the report does not say whether the two overlap.

Gaps in the record

The page says the incident is “part of a broader AI-powered attack campaign” reported in late October, based on shared tooling and techniques. It does not say what, if anything, this attacker took beyond what is listed, and it gives no customer count. The page header shows an update of November 11, one day after the November 10 note, with no explanation. A KuCoin news item relaying a BlockBeats flash report presented the technical details on November 12 as fresh; they were on Chainstack’s page two days earlier.

The Take

“Third-party component” is accurate and tells a customer almost nothing. The facts that matter are in the technical section: a default password, no login layer, an agent tool server with cluster-admin and no authentication. Chainstack deserves credit for publishing indicators and for saying plainly that no vulnerability was exploited. Infrastructure vendors that hand out node credentials should lead their notices with that kind of detail, not bury it a week later.

More on the subject